Research On Blind Deconvolution Methods Based On Retinex Operator

Image restoration is an important branch in the field of digital image processing.To establish the degraded model based on the knowledge of image degradation and process the degraded image from the objective views,makes the image clearer as much as possible.Computational imaging is making use of mathematical model to describe accurately and analyzing some system decay and error model further based on the process of optical imaging,then through the calculation method to reconstruct the image.Image degradation is the decline in image quality affected by the hardware system and the impact of the external environment.Some important information about the latent image is lost and cannot meet people’s applications.As a result,there is a demand to restore the degraded image and obtain a clear image closer to the real image to help people get the lost important information meanwhile;it also helps the upper image analysis and understanding.Image restoration are of great significance in the fields of astronomy,medicine and military and other aspects.However,the Point Spread Function that causes image degradation is often impossible to know accurately in practice.Therefore,it is a great meaningful to study the method of image blind restoration,which also called blind deconvolution.In this paper,this study is mainly how to get the real point spread function and recover the original image from the blurry image whose prior information is unknown or partially known to.We have researched on the common noise model,PSF model,image quality evaluation,the basic theory and common techniques of image restoration from the traditional frequency domain filtering method to the subsequent regularization technology,partial differential equations,and the recent neural network technology.At the same time,the common methods are studied to solve the image blind restoration problem,like primal-dual,Iteration and alternating direction method of multipliers,and then master the meaning of the degradation model and the core how to solve the restoration problem.This article has been improved from two aspects both how to get more real PSF and how to find more effective regularization,and proposed a method of solving PSF based on multiple blurry images in the same scene and an image restoration method based on regular constraints of Retinex operator.The main innovations are as follows:(1)There is a detail study about how to estimate the PSF for image blindness restoration.The information provided by a single blurry image is not sufficient,because of that a new method to better estimate PSF is proposed based on multiple blurry images of the same scene.First,pr-estimating the PSF and then take advantage of the share PSFs of two images for PSF constraints,and add up to cross regularization term and with the normalized and minimize the partial derivative L~p norm.The experiment results has proved that the PSF obtained by the improved model has better effect and closer to the real PSF which can recover blurry image better.(2)On the basis of the total variation,this paper takes advantage of the inner correlation which constrain to each other between the three channels.A new regularization term is proposed based on Retinex operator which makes full use of image prior to help the other two blurry channels recover better,The experimental results show that the proposed algorithm in this paper has a better works on the images whose single channel is clear and the other two channels are not clear.